Skill assessment of environmental parameter on cyclone intensities.

“

Assessing the performance of the statistical and l models predicting the cyclone intensity is the need the day. . However, it is difficult to compare these models across TC basins, as these models have been developed independently, using different predictors, atmospheric datasets and training periods. In this paper, we develop statistical–dynamical linear hindcast models for each TC-prone basin using the same set of predictors, atmospheric dataset using extended training/validation period (1979–2012). We used this common framework to assess the relative importance of TC initial characteristics versus environmental parameters in the models’ skill, identified the most skilful environmental predictors in each basin, and investigated the sensitivity of the skill to the TC strength. Our results indeed reveal that a model built from climatological environmental parameters along the TC track yields nearly the same predictive skill as a model built from real-time values. For implementation purposes, it is indeed far simpler to use atmospheric predictors calculated from their seasonal climatology than to retrieve the real-time forecasted values of those predictors along the forecasted TC track. However this strategy may not be adapted to the Southern Hemisphere, where our results suggest that skill gains from accounting for real-time environmental parameter variations.

”